Functional languages do away with the current state paradigm and achieve referential transparency. They also exhibit inherent parallelism. These qualities fit very well on top of a data-driven architecture such as a data flow machine. In this paper, we propose a fully reconfigurable data flow machine for implementing functional programming languages. The design is based on smart memories and nodes interconnected via a hypercube. Important aspects of the proposed model are described and compared with other similar attempts. Advantages of our system include massive parallelism, reconfigurability, and amenability to higher-level, graphical programming. Current limitations are identified and extensions are suggested.